Public Trust in the AI Era: A Dialogue with Terry Flew

Terry Flew is Professor of Digital Communication & Culture, Faculty of Arts & Social Sciences at the University of Sydney and an Australian Research Council (ARC) Laureate Fellow. In 2024, he commenced a 5-year ARC Laureate Fellowship for a 5-year project on Mediated Trust: ideas, Interests, Institutions, Futures. He is also Co-Director of the Centre for AI, Trust and Governance (CAITG) at the University of Sydney.

Yao Yao:

In your recent publications, you have described trust as an invisible institution. Why and what have been your main observations over the years? What phenomena and social changes have inspired you to study trust?

Terry Flew:

The idea of trust as an invisible institution comes from the economist Kenneth Arrow. In his book The Limits of Organization, he observed that trust is not a commodity, and it cannot be bought or sold, but it is absolutely central to the functioning of market economies (Arrow, 1974). This draws on an insight that actually goes back to the 18th century and to the key figures of the Scottish Enlightenment, who saw that there was this space between the state and the market, the space of civil society, in which the relationships between people were essential to the effective functioning of society and even the effective functioning of the economy.

So the question of trust is one of those things that is hard to measure. It’s difficult to know if it exists, and it is not particularly notable when you do have it. Trust tends to be talked about when it’s declining or when it has disappeared. An author named Geoffrey Hosking has described trust as being like a coconut tree, which takes a long time to grow but is quite easy to chop down. And once you chop it down, it will be very hard to make it grow again (Hosking, 2014). In this sense, in addition to the formal social institutions of law, government, education, health care and so on, we have these invisible institutions, this sort of cultural milieu through which trust is maintained. The question has been studied more intensively over the past 20 years, in part because of the perception that trust has declined, at least in Western countries. Therefore, people began to pay more attention to issues such as what trust is, what the conditions for trust are, and what factors lead to the loss of trust.

In terms of wider societal changes, we have seen significant developments in a number of countries since the beginning of the 21st century. For instance, we know that the decision to go to war in Iraq was based on false information about weapons of mass destruction, which engendered a loss of trust in the United States, the United Kingdom, Australia and on other countries. The 2008 global financial crisis caused people to fall into debt, and trust in financial institutions declined.

There is also the question of the impact of social media on other institutions, particularly how the rise of social media relates to the decline of traditional media. To what extent are the values upheld by traditional journalism no longer applicable in the social media world? Will this lead to a loss of trust? Is there also a loss of information overall? Meanwhile, issues such as social inequality, political polarization and loss of social cohesion were also discussed as factors that catalyse a ‘crisis of grievance’, that manifests itself both on social media and in real life.

Yao Yao:

Based on the term ‘mediated trust’ that you have developed in your work, can we say that communication is a catalyst for changes in trust?

Terry Flew:

Yes. One of the arguments I made in my 2020 International Communication Association (ICA) Presidential Address was that communication as a field actually has quite a bit to say about trust, but the term is not used as much in communication as compared to sociology (Flew, 2021a). In sociology, trust is one of the foundational concepts, because sociology starts from the premise of how we can live together as a society without fear. So, the question of trust has always been there. However, I think trust has been an implicit concept in communication. I think the field of communication might be more concerned with truth, or to put it another way, with what I call the systematic reproduction of mistruth or ideological theories. We pay a lot of attention to whether or not truth is represented or misrepresented in communication.

But one of the things we know about communication, and even the most rationalist scholars like Jürgen Habermas have acknowledged, is that the relationship between the speaker and the receiver is not just an epistemic one, nor is it just about the accuracy of the information being conveyed. It’s also about empathy, the speaker’s ethics and sense of purpose, and whether they share the same feeling with the listener. So there’s an emotional dimension as well, and I think that becomes important to trust.

The other point to make about mediated trust is how other authors have used this term. Balázs Bodó from the University of Amsterdam, Mike Schäfer from the University of Zurich, and, more recently, Jayson Harsin from the American University of Paris have used the term. The fact is, we are increasingly connected to each other through technology. Therefore, this mediation does not simply refer to the media itself or the media industries themselves but rather to the broader sense of mediatization that European communication scholars have specifically talked about, that is, the extent to which the communication technologies we use to interact shape our relationships with each other. The question is whether we trust each other, and our relationships are always enhanced through technology as well as face-to-face interactions.

Yao Yao:

Besides the use of media technologies, are there other trends that are leading to changes in trust?

Terry Flew:

One of the questions that comes up is whether you are developing a media-centric account of these questions or a non-media-centric account. What I mean by that is how important media is or how important communication is to the changes that you are talking about. And communication scholars will tend to say it’s very important, while political scientists or sociologists or others might talk about something else.

One example is the work that the political economist Thomas Piketty and his colleagues have done on rising inequality. Their recent study identified rising inequality across 50 countries over the last four to five decades and the extent to which that is connected to declining trust. More importantly, this is the challenge that Piketty’s most recent book Capital and Ideology presented, which is if you’ve got this rising inequality under capitalism, why didn’t you get a rise in anti-capitalist movements? Why are the major political beneficiaries, at least in the West, populist parties and movements? Why, then, is discontent with growing inequality in the United States tied to the rise of someone like Donald Trump, who is obviously a billionaire? Piketty’s point is that inequality in income and wealth is also accompanied by another kind of shift: a shift in the possibilities between the more highly educated and the less highly educated (Piketty, 2020).

In the United States, most of the benefits of economic growth over the past two or three decades have been captured by so-called coastal elites, especially at elite universities. While a lot of the job losses are impacting on blue-collar workers in the old industrial heartlands, the question of mediated trust there is who can speak up to quell that discontent. To some extent, Trump’s concerns are justified, even if his solutions are unlikely to address the problems. He has actually been more effective in speaking to that working-class discontent than the Democrats have. Whether he gave the right answer is a completely different question, but in a sense, that kind of angry communication style, if you will, ‘they’re ripping us off’, has worked. This approach resonates in a way that more rationalist approaches do not. So you have the sense that there are clearly other factors at play besides the extent to which communication technologies are changing the way we build trust with unseen others.

Yao Yao:

You have been following the ‘policy turn’ towards digital platforms for years, as shown in your book Regulating Platforms. How do you think this relates to the crisis of trust in digital communications?

Terry Flew:

Broadly speaking, Regulating Platforms attempts to capture a kind of ‘synoptic history’ with three phases. The first phase was the open internet, which emerged from the grassroots and challenged traditional hierarchies. We have been committed to a philosophy of openness since the late 2000s. What is clear is that what we call the internet is increasingly becoming a series of large digital platforms through which most communication flows take place. And these platforms are owned and controlled by the biggest companies in the world. So the open internet was changing to what I referred to in that book as the platformization of the internet. The policy turn is a third stage where the contradictions of these developments become more apparent. Citizens increasingly look to governments to regulate the giant platforms, and we can see this confrontation between governments and large platform companies around the world, including in China (Flew, 2021b).

About whether this situation will continue, I think we are at a fork in the road right now as the situation develops in the United States, with the Trump administration removing most of the Biden-era regulations. Faced with this real change, platform companies are lobbying the rest of the world to remove regulation, as is most evident in the current dispute between the United States and the European Union. Therefore, I think whether the policy shift will persist will depend in part on how this dispute plays out.

At the same time, I also think that the issues that are presented by digital platforms and social media are not disappearing. In Nick Couldry’s newest book, The Space of the World: Can Human Solidarity Survive Social Media and What If It Can’t, he argues that a series of industry and policy decisions over the past 30 years have culminated in a communication dilemma that, in his view, is so severe that we have a large number of uncontrolled platforms that are driven by commercial interests and have developed highly manipulative and addictive algorithms (Couldry, 2024). This in turn has triggered social polarization, alienation and so forth. It is hard for national governments to intervene because most of these platforms are not based in their own countries.

Furthermore, we are now seeing a growing willingness to use international trade instruments to reduce the regulation of platforms. In addition, the companies themselves have periodically taken steps to try to take action on content on their platforms. For example, Meta has established an Oversight Board; Twitter used to have a Trust and Safety Council. They now feel empowered to completely roll back all of those measures. But then again, you will also find that the reasons that these entities were initially created to deal with, such as misinformation, fake news, discrimination and vilification, will still exist. So I don’t think these issues have been ultimately resolved.

Yao Yao:

In your opinion, artificial intelligence (AI) will be the next frontier of debates around trust in communication and media. What has happened and why?

Terry Flew:

The concept of AI has been around since World War II. Alan Turing proposed the concept of thinking machines in 1950. In 1956, the first AI conference was held at Dartmouth College in the United States. AI has a long history and covers various fields, from robotics to machine learning. AI has long been a big part of popular culture, with films such as Stanley Kubrick’s 1968 film 2001: A Space Odyssey or the replicants in Ridley Scott’s 1983 film Blade Runner. While filming 2001, Kubrick was actually in conversation with Marvin Minsky, a professor of AI at MIT. In 2001, we see how the computers on the spacecraft navigate space, how they understand what humans want to do and how they control the spacecraft. This is what we would today call Artificial General Intelligence (AGI).

In the early 2020s, with the emergence of ChatGPT and other technologies, people quickly realized that AI could replicate human thinking to a certain extent. For instance, in college, students can use AI to write a paper, and the AI can write a pretty good paper for them. This raises a lot of questions about how to grade student papers, because when you grade the paper, you’re also grading the student. But if the content isn’t student-generated, what are you evaluating? Are you evaluating the AI? That’s a different question from whether you’re evaluating the student. This speaks to some of the debates around whether we can trust images or the written word. What does it mean to trust AI? What does it mean to communicate with a chatbot? What kinds of creativity can computers unleash? Will AI increasingly create new stories? Is that always a good thing or a bad thing? There are so many questions.

In a sense, we are dependent on AI. If I continue with the student essay example, one of the challenges with AI is that there might be AI detection software. How effective is AI detection software at detecting AI? If AI detection software is used to detect AI, allowing two groups of machines to talk to each other, this could be a good thing. We can forget about grading. Let the machine grade the papers, and then everyone can go to the park and chat about the meaning of life. That’s one possibility. But you can see that trust operates on multiple levels, and it’s very different from universities’ traditional concerns about plagiarism or what is known in Australia as contract cheating. These are all issues. Contract cheating is obviously wrong. It is also wrong to pay someone to write a paper for you or to use a software program to do it for you. The paper itself is in a grey area. More importantly, a great deal of the communication we currently have in the public sphere will increasingly be generated by AI, and this raises a new set of questions about trust.

Yao Yao:

‘Trust in AI’ can be interpreted from many perspectives, which have also caused some conceptual confusion. How do you understand this concept?

Terry Flew:

First, computer scientists come to the trust question quite differently. When computer scientists talk about trusted systems, they are referring to the resilience of the system, that is, the extent to which the computer system can prevent being hacked or content can be stolen. For example, Apple computers are known for being trustworthy because they are difficult to hack into. Therefore, the concept of a trusted system is essentially a security concept.

The sociological conception of trust is quite different. It’s about authenticity. The word ‘confidence’ appears frequently in these discussions. German sociologist Niklas Luhmann once pointed out that without trust, you can’t get out of bed in the morning and you can’t do anything (Luhmann, 1979). You certainly can’t fly from Australia to China like I did, because you don’t trust the pilots, the engineers and everything they did. These are myriad layers of trust. Any complex social activity must rest upon an implicit notion of trust, or confidence that others are doing their jobs properly. So trust and confidence are closely related. For example, I don’t need to know anything about pilots to trust that the plane can fly, and that’s a good thing. Because if I had to know everyone I came into contact with before I could trust them, I’d have to hide in a cave.

So I think trust in AI is partly about whether we can trust its outputs, whether they provide legitimate information, and whether we trust that the large companies that develop AI systems are trustworthy, or at least ethical in their practices. There is a lot of discussion about the inputs to AI and whether data harvesting is ethical or illegal. Therefore, these issues are also about whether technology can be used to build trust.

To illustrate, blockchain has been promoted as a trust technology. I know some people who are involved in using blockchain to export Australian beef to China. One of the challenges there is, how do you know it’s Australian beef? Because someone put a little Australian flag on the plastic wrapping and said, well, it’s Australian. Anyone could do that. In fact, the idea behind blockchain is that the proof of concept that you initially register cannot be changed during the transaction. Once something is registered on the blockchain, it’s immutable. So blockchain is seen as a trust technology that can enable things like smart contracts and thus enhance trust in the sense of confidence in economic transactions.

In summary, trust in AI involves different levels: at the company level, at the software engineer level, and at the level of how the data is acquired. The question is whether the information generated by AI can be biased by where the data originally came from. We also face the question of whether we know if something is generated by AI and whether that matters. When Dwayne ‘The Rock’ Johnson saved San Francisco in the movie San Andreas, we didn’t spend much time worrying about whether it looked authentic. We are used to a certain degree of image manipulation. We tend to distinguish movies, entertainment, games or anything else from real life. Now, however, with real-life information increasingly being machine-generated, it has introduced a whole new layer of trust.

Yao Yao:

In your research, you mentioned that people’s trust in AI would affect people’s trust between each other. Could you explain why?

Terry Flew:

We are currently experiencing a wave of conspiracy theories. Some of this has to do with knowing that machines are incredibly powerful. We want to know what is real and what is behind this. For example, when a flood or other catastrophic situation occurs, people may ask, is someone using machines to control the weather? The current surge in conspiracy theories is happening at a time of great uncertainty, when people are aware of the power of machines and encounter a certain level of suspicion about many things. And at different times in history, periods of rapid change and uncertainty have led various people to become more suspicious of others because they worry about what might be going on behind the scenes.

There are two broad approaches to this problem. One is to return to Thomas Hobbes’s Leviathan theory, which states that humans don’t trust each other and, if left to their own devices, will kill each other. So you need a powerful state to try and control this impulse to domination. Historically, that’s inherently problematic. Another view is that humans can be kind to each other. But you need conditions that foster a certain degree of altruism. And that’s what government can foster. It requires people to be willing to listen to one another, to learn to engage in civic dialogue and to express themselves. If these conditions can be created, then some of the positive possibilities of AI really begin to emerge. So, are governments around the world taking steps to foster these conditions? That’s something worth noting.

Yao Yao:

The advancement of AI not only involves tech giants but also reflects complex geopolitical interconnections, such as the competition between China and the United States. From this perspective, how do you believe geopolitics might shape public trust in AI?

Terry Flew:

It has long been obvious that the technological competition of our time is primarily between the United States and China. According to statistics, more than 80% of the world’s AI patents are generated by China and the United States, of which China accounts for about 60% and the United States accounts for 20% (Maslej et al., 2024). The rest of the world is primarily the recipient or regulator of these systems. But the largest companies in this field are, without a doubt, headquartered in the United States or China. So, as each tech company vies for dominance in the technological space, they are well-positioned to leverage the influence of their respective governments to promote themselves as national champions, building connections and vying for influence through foreign policy and other channels.

To be specific, Australia is located in the South Pacific, which has many small island countries and some developing countries, such as Papua New Guinea. What’s happening in these regions is that when Australia withdraws from these countries, China will often take the initiative. And when China enters the market, the United States may intervene. Then there may be an expectation that Australia build the local telecommunications system. Of course, leaders in the Pacific region are well aware of this. They are ready to engage in dialogue with all parties.

In the digital field, the United States has clearly long dominated. China is developing rapidly in this area and is particularly influential in the Global South. But in the field of AI, the competition for data acquisition is ongoing. I think over the last decade, governments have increasingly viewed each other as competitors. Confidence in the multilateral order is declining among governments, which are increasingly concerned about their own sovereignty and resilience. This has significantly impacted public trust in AI.

However, the success of DeepSeek has shown us that it is possible to build highly robust and efficient AI systems without large amounts of data. In our past cognition, the premise of the data war is that whoever has the most data will win. So all these companies are pulling in data from everywhere. Data are often a competition between companies, and some companies have a natural advantage in this area. For example, Google, as the world’s leading search engine, has accumulated a large amount of data over its more than twenty years of development. This is advantageous in the field of AI. Nevertheless, more advanced algorithms are able to produce results of a higher standard without the same amount of data. Then it’s possible that the largest companies won’t necessarily dominate the field. I think we’re going to see a lot of challenges and changes. We may see a reversal of the past 15–20 years of software training. I’m not sure, but geopolitics will certainly have a huge impact on public trust in AI.

Yao Yao:

To enhance public trust in AI and communication technologies, what policies and practices do you think should be formulated? And who should be the actors in establishing these policies?

Terry Flew:

I think the major concepts are around ethics, risk and bias. The question of ethics refers to the extent to which a company is bound by some statute or regulation, whatever it may be, that does not allow its products to be used in ways that cause harm.

Regarding the issue of risk, it can be broadly categorized as minor, medium and high. Minor risks won’t cause any serious harm, such as translation software. If the translation results appear ridiculous, that’s just silly, but no one will die. High risks refer to areas where human life is at risk, such as AI drivers and cars running over people or medical robots monitoring surgeries. People could potentially die from these AI applications. The EU AI Act has placed a lot of guardrails around high-risk areas. The United States began moving towards these safeguards during the Biden administration, but the Executive Order instituting these changes was overturned by Trump.

The intermediate or medium-risk area is interesting, and its significance can be underestimated. AI is increasingly being used for things like job selection, loan decisions and so on. These decisions don’t look as bad as a self-driving car hitting someone, but they do affect people’s lives, possibly more than the high-risk concerns.

This is where the problem of bias comes in, because of the way data are often collected. In many cases, the people most likely to be affected by AI decisions are often the ones least likely to have contributed to the dataset. Just as we would condemn racially discriminatory hiring decisions, if an AI system produces similar results, it becomes a human rights issue and a societal issue. This requires government involvement.

Recent experience has shown that companies are prepared to ignore issues and forego corporate social responsibility. They are also prepared to pay fines as the cost of doing commercial business. Therefore, to address these issues, governments must be involved. But again, we face the classic problem of how national heads of government can use global platforms to take action in the absence of some kind of global governance regime. I think these issues are likely to persist.

Yao Yao:

Given the rapid development of AI, could you give us an outlook on the future direction and trends of communication research? How do you view the new meaning of Communication and the Public?

Terry Flew:

Journals such as Communication and the Public have been actively focusing on the internet and social media sectors. We are now entering a new phase in our interaction with AI. We will face new challenges, including how to build an interdisciplinary academic model, how to understand human-machine interaction, what we can learn from fields such as computer science and engineering, and how to apply these experiences to our future academic research. It’s important to remember that many of the early founders of communications, such as Claude Shannon, were themselves engineers involved in early AI research. Therefore, while we maintain critical thinking, we also need to pay attention to how these technologies work and their social impacts.